Opportunities for evaluation using the
Stepped Wedge trial design
Celia Brown, Alan Girling, Prakash Patil and Richard Lilford
Department of Public Health and Epidemiology
Today’s presentation




Describe the stepped wedge design
Detail when the design might be useful
Consider the advantages and
disadvantages of the design
Review 12 studies employing a stepped
wedge design




One individual/cluster
receives the
intervention in each
time period
Order of intervention
determined at random
All individuals/clusters
get the intervention by
the end of the process
Data collected in each
time period
Participants/Clusters
The Stepped Wedge Design
5
4
3
2
1
1
2
3
4
Time periods
Shaded cells represent intervention periods
Blank cells represent control periods
Each cell represents a data collection point
5
6
When is the design useful?



Prior belief that the intervention will do
more good than harm – ethics of
exclusion
Logistical, practical or financial constraints
to simultaneous intervention
Evaluating a public policy intervention that
is being rolled-out before effectiveness
demonstrated (e.g. Sure Start)
Advantages



Enables RCT approach in situations where
parallel design not possible
Can model the effect of time of
intervention on effectiveness
Can model the effect of length of
intervention on effectiveness
Disadvantages



Requires extensive data collection, so best
where routine data are to be used
Additional time analyses only appropriate
if no cluster effect or cluster x time
interactions
Currently no published guide to data
analysis (but watch this space!)
Review of Stepped Wedge studies

Comprehensive literature search found only 12 papers
or protocols:
Lead Author
Gambia Hepatitis
Study Group
Cook
Wilmink
Date
1987
Disease
Liver cancer
Country
Gambia
Setting
Regions
1996
1999
USA
UK
Workplace
GP surgeries
Somerville
Fairley
2002
2003
UK
Australia
Houses in Watcombe
Sexual health clinic
Hughes
2003
2004
Zambia and
Uganda
Australia
Health clinics
Levy
Priestly
Bailey
Grant
Ciliberto
2004
2004
2005
2005
Substance abuse
Ruptured abdominal
aortic aneurysms
Respiratory Health
HIV (Adherence to
antiretroviral therapy)
HIV (Mother to child
transmission)
HIV (Adherence to
antiretroviral therapy)
Critical care
Water-borne diseases
TB in HIV+ men
Childhood malnutrition
Chaisson
2005
TB in HIV+ men
Brazil
UK
South Africa
South Africa
Malawi
Ambulatory care clinic
in a tertiary hospital
NHS hospital trust
Households
Company health centre
National rehabilitation
units
HIV clinics
Randomisation and Sample Size
Author
Level of stepping
Randomised?
No. Steps
Gambia Hepatitis
Study Group
Cook
Wilmink
Vaccination team
Yes
17
Cohort
Individual
Yes
Yes
2
13,147
Somerville
Fairley
Hughes
Levy
Priestly
Bailey
Grant
Ciliberto
Chaisson
Sets of houses
Not stated
Pre-natal clinic
Individual
Ward
District
Individual
Rehab unit
Clinic
Yes
Yes
Not stated
Yes
Yes – in pairs
Not stated
Yes
Not stated
Yes
2
43
2
Not stated
8
4
1,655
7
29
No. Participants
Intervention
Control
61,065
63,512
Yes
No
371
29,713 person
years
SS Calc reported?
70,298 person
years
119
43
Aim: 304
Aim: 304
68
2,903
4,547
400
1,655
992
186
Not stated
No
No
No
Yes
No
Yes
No
No
Yes
No
Reported motivations






Ethical (n=4)
Practical problems of simultaneous intervention
(n=4): insufficient resources (n=3); logistical
difficulties (n=2)
Maintain RCT for evaluation (n=4)
Detect underlying trends/control for time (n=4)
Individuals/clusters act as own controls (n=2)
None (n=1)
Methods of Data Analysis
Lead Author
Gambia
Hepatitis
Study Group
Cook
Wilmink
Somerville
Fairley
Hughes
Levy
Priestly
Primary outcome
measure
Liver cancer
rates/Vaccine
efficacy
Health Behaviour
Questionnaire
measures
Incidence and
mortality of RAAAs
Respiratory Health
Symptoms
Proportion of
missed doses
Mother to child HIV
transmission
Proportion of
missed doses
Rate of in-hospital
deaths
Bailey
Water quality
Grant
TB episodes >90
days after clinic
entry
Attainment WHZ
score >-2/Death
Ciliberto
Chaisson
TB Incidence
Method(s) of Analysis
Comparisons of incidence rates on a step
by step basis to identify vaccine efficacy
Comparison of group means and group by
time, gender and education interactions (Ftest)
Poisson likelihood distribution for incidence
rates in person years and maximum
likelihood rate ratios
Not stated (description of intervention only)
Unpaired t-test of means
Not stated (protocol only)
Wilcoxon rank-sum test
Logistic regression
Cox proportional hazard models (length of
stay)
Summary statistics only
Time series analysis for diarrhoea rates
Poisson random effects model
95% CI for differences between groups
Linear and logistic regression for effects of
covariates
Step by step analysis of incidence
Conditional logistic regression
Cost-effectiveness analysis
Conclusions: Design



Stepped wedge design has significant
potential for evaluating public policy
interventions using a RCT
Intensive data collection means design
most appropriate where routine data used
Opportunities for assessing different
effects of time
Conclusions: Review




Review highlighted dearth of evaluations using
the stepped wedge design
Variety of interventions and settings establishes
design’s potential
Need to ensure studies reported to same
standards as other trials (e.g. CONSORT) –
particularly sample size calculations
Variety of statistical approaches to data analysis
implies need for standardised approach
Questions?
Celia Brown: c.a.brown@bham.ac.uk
0121 414 6043
3" Wide Stepped Wedge